AI GOVERNANCE

ISSUE 01 AI GOVERNANCE

Using ISO 55000 to govern AI in your Maximo estate

An AI without context can do something plausible but operationally wrong. The context it needs already exists, in the management system you run to meet ISO 55000. Here is how that applies to Maximo.

Alex Delic/ 23 October 2026/ 5 min read

From the MaxTAF newsletter, issue 1, 23 October 2026.

The piece in four minutes. Narration by a synthetic voice; script by the author.

Give an AI a task without context and it can do something plausible but operationally wrong. Context is what turns a record into a decision: which assets matter, what the rules are, what risk you are prepared to carry. Traditionally, software never had to hold any of that, because people did. The planner knew why pump 3 was critical and the maintainer knew which jobs could wait, so the system only ever needed to hold the record.

We now expect an AI to work at that same level. Without the context written down where it can read it, it cannot. For most operators that context already exists, in the management system they run to meet a standard like ISO 55000: the criteria, the strategies, the approval rules. Written for auditors, and now the brief an AI needs. This piece is about how that applies to Maximo.

Maximo is the record layer of an ISO 55000 asset management system. The decisions live above it. An AI connected to Maximo can see that an asset carries priority 1 and who may approve the work, but not the criticality criteria that set that priority, the strategy for that asset class, or the risk you have decided to accept. Before it gets any authority, it needs that layer.

What ISO 55000 asks you to write down

The standard does not hold your context; it asks you to hold it. ISO 55001 asks each organisation to write its intent down in six places, and if you are certified they already exist:

  • Line of sight: how each piece of work traces back to your objectives, through the strategic asset management plan (SAMP).
  • Criticality and risk criteria: which assets matter, and what risk is acceptable.
  • Maintenance strategy: how each asset class is to be maintained.
  • Roles and authorities: who may decide what, and the approval thresholds that apply.
  • Performance measures: how you know the system is working.
  • Data and information: what your decisions depend on (clause 7.6).

Ofwat's 2028 licence condition requires water companies to prove a fit-for-purpose asset management system, and ISO certification is one route. The Office of Rail and Road required each Network Rail route to show ISO 55001 alignment.

An organisation can run Maximo well and still fail a 55001 audit.

What Maximo holds, and what it does not

Maximo holds the register, the work, the costs, the history, a priority on each asset and an approval limit on each user. It does not hold the reasoning behind them, which is what an AI needs to apply them to a new case.

On one offshore drilling operator's Maximo, a custom status marked an asset as subsea, and subsea carried a 180-day maintenance trigger. The status was in the record; the trigger, and a three-part criticality of environmental, operational and safety, were in people's heads. An agent reading the record could see the status and nothing else.

Why the bare record is not enough

An agent over the record alone is not wrong by default. It is wrong wherever your policy differs from the pattern in the data, and it has no way to know where that is. Maximo Application Suite 9.2 lets you connect your own agents to Maximo Manage, so this now matters. The record holds the outcomes of past decisions, not the rules that produced them, so where it is silent the agent infers from the pattern: pumps like this one usually get priority 1, so it drafts the work order that way. Most of the time that is fine. But this asset sits under a run-to-failure strategy, and nothing in the record says so. The draft is plausible but operationally wrong, exactly where a maintainer would have caught it.

Three moves, in order

In this order: you cannot read the record against criteria you have not written, or audit data against requirements you have not set.

01

Write it down

Locate the decision layer, or write it. If you are 55001-certified, most of it already exists in your asset management system; if not, the six items above are the minimum.

02

Read the record against your policy

Let the AI read the record inside that layer, read-only first. "Raise a job on pump 3" should return a priority from your criticality criteria, a work type from your strategy, and a flag when the request contradicts your policy. A person approves anything that writes. The policy has to sit where the agent can read it, alongside the records.

03

Audit the data

55001 makes the fitness of asset information auditable. Point the same read-only connection at your criticality register and your data requirements, and find the blanks before your auditor does.

It is not only 55000

ISO 55001 is one of a family of management-system standards, and the system behind each is your written intent for a different part of the business. Whichever standards you hold, the management systems behind them are context an AI can be given. The system behind ISO 27001 already says who may access what and how a change is approved. ISO 9001 says how work is checked before release; ISO 14001 and ISO 45001 hold the environmental and safety criteria inside a criticality score. None was written with an AI in mind, but the intent is documented and audited, so the agent inherits it instead of guessing.

What this means for your AI position

Governing AI in maintenance uses the system you already have: the same line of sight, criteria, roles and approval thresholds, applied to a new kind of actor, and explained to an auditor in terms the auditor already uses.

The management system you run for ISO 55001 already holds that intent, and organisations that have written it down will see AI pay off fastest.

Start with whichever of the six is thinnest. That is where an AI goes wrong first, and where the fix is cheapest.

See how we put AI to work inside a Maximo estate, read-only first, within the policy the organisation already wrote down.

SUBSCRIBE ONE ISSUE A MONTH

Get the MaxTAF newsletter by email.

A short cover note on the first Tuesday of the month, with a line on each piece and a link back here. No tracking pixel, and you can stop it with one reply.

Or email news@maxtaf.com with the word subscribe.